| Message ID | CAFk3UF_e+653MEPr4YR=tdKX=969WnNHt+0640NCUUk-yT+EdA@mail.gmail.com |
|---|---|
| State | New |
| Headers | show |
On Fri, 31 Mar 2017, Sebastian Pop wrote: > On Fri, Mar 31, 2017 at 1:50 PM, Richard Biener <rguenther@suse.de> wrote: > > That looks like a workaround rather than the correct fix. Who sets .MEM? It should assign the proper def. > > > > Moving the auto_profile pass above several other passes fixes the problem. > The minimal change to fix the problem is by disabling these two passes: > > --- a/gcc/passes.def > +++ b/gcc/passes.def > @@ -86,14 +86,12 @@ along with GCC; see the file COPYING3. If not see > NEXT_PASS (pass_ccp, false /* nonzero_p */); > /* After CCP we rewrite no longer addressed locals into SSA > form if possible. */ > - NEXT_PASS (pass_forwprop); > NEXT_PASS (pass_early_thread_jumps); > NEXT_PASS (pass_sra_early); > /* pass_build_ealias is a dummy pass that ensures that we > execute TODO_rebuild_alias at this point. */ > NEXT_PASS (pass_build_ealias); > NEXT_PASS (pass_fre); > - NEXT_PASS (pass_early_vrp); > NEXT_PASS (pass_merge_phi); > NEXT_PASS (pass_dse); > NEXT_PASS (pass_cd_dce); > > It seems to me that early_vrp is the one changing the IR in an inconsistent way. No, that would have bombed in verify_ssa.
--- a/gcc/passes.def +++ b/gcc/passes.def @@ -86,14 +86,12 @@ along with GCC; see the file COPYING3. If not see NEXT_PASS (pass_ccp, false /* nonzero_p */); /* After CCP we rewrite no longer addressed locals into SSA form if possible. */ - NEXT_PASS (pass_forwprop); NEXT_PASS (pass_early_thread_jumps); NEXT_PASS (pass_sra_early); /* pass_build_ealias is a dummy pass that ensures that we execute TODO_rebuild_alias at this point. */ NEXT_PASS (pass_build_ealias); NEXT_PASS (pass_fre); - NEXT_PASS (pass_early_vrp); NEXT_PASS (pass_merge_phi); NEXT_PASS (pass_dse); NEXT_PASS (pass_cd_dce);